PA ML Benchmark Report

Class Balance & Missingness

Class Distribution

Missingness per Column

Feature Correlations

Numeric Distributions

Distributions by Class

Table 1 — Cohort Characteristics

Feature	Type	Overall_mean	Overall_sd	Class0_mean	Class0_sd	Class1_mean	Class1_sd
Haemoglobin	numeric	12.241	2.181	13.983	1.228	10.499	1.392
Haematocrit	numeric	37.12	6.029	42.015	2.941	32.225	4.014
MCV	numeric	96.982	9.078	89.705	4.758	104.26	6.018
MCH	numeric	31.992	2.839	29.973	2.108	34.01	1.878
MCHC	numeric	32.997	1.364	33.951	1.01	32.044	0.94
RDW	numeric	15.038	2.55	13.039	1.036	17.036	1.986
WBC	numeric	5.269	1.862	6.575	1.419	3.963	1.229
Platelets	numeric	212.64	66.215	247.397	48.392	177.884	63.364
Serum_B12	numeric	281.397	149.936	399.019	120.709	163.774	52.063
Folate	numeric	8.571	2.84	8.985	3.034	8.157	2.567
Methylmalonic_Acid	numeric	474.957	249.058	251.13	60.481	698.784	141.873
Homocysteine	numeric	20.039	10.361	12.055	4.29	28.023	8.293

Showing first 12 rows. Full CSV: eda_table1.csv

Group Comparisons vs. Target

Feature	Type	Test	Statistic	PValue
Methylmalonic_Acid	numeric	Welch t	-64.90347220212456	2.800092021824305e-292
Haematocrit	numeric	Welch t	43.993063238484176	5.569903702600216e-228
Haemoglobin	numeric	Welch t	41.96262858268173	2.5578108938549414e-221
MCV	numeric	Welch t	-42.42145333046377	3.0982816336581425e-221
LDH	numeric	Welch t	-42.48169063174602	3.6385036972064432e-199
RDW	numeric	Welch t	-39.90085154659209	8.047338959245091e-188
Serum_B12	numeric	Welch t	40.01468054300914	1.0307855038226757e-180
Homocysteine	numeric	Welch t	-38.24118589203447	3.499258365804435e-178
MCH	numeric	Welch t	-31.98128704487607	1.6926867353276502e-154
WBC	numeric	Welch t	31.11813173711018	2.5155083314066547e-148
MCHC	numeric	Welch t	30.904618123761804	1.7963876643700358e-147
Bilirubin	numeric	Welch t	-23.682244363192503	7.412536433946976e-93

Showing first 12 rows. Full CSV: eda_group_tests.csv

Grid Search Diagnostics

$Grid Search Overview$

Overview of per-model grid-search panels (log loss; lower is better). Shaded bands indicate 95% CI across settings of other hyperparameters.

LOGISTIC — Grid Search (log loss)

Raw CV results: gridsearch_results_logistic.csv

RF — Grid Search (log loss)

Raw CV results: gridsearch_results_rf.csv

XGB — Grid Search (log loss)

Raw CV results: gridsearch_results_xgb.csv

SVM — Grid Search (log loss)

Raw CV results: gridsearch_results_svm.csv

Model	LogLoss (↓)	AUC	Accuracy	Precision	Recall	Best CV LogLoss
LOGISTIC	0.0019	1.000	1.000	1.000	1.000	0.0009
XGB	0.0042	1.000	1.000	1.000	1.000	0.0046
SVM	0.0074	1.000	1.000	1.000	1.000	0.0071
RF	0.0155	1.000	1.000	1.000	1.000	0.0182

Model

LogLoss (↓)

AUC

Accuracy

Precision

Recall

Best CV LogLoss

LOGISTIC

0.0019

1.000

0.0009

XGB

0.0042

1.000

0.0046

SVM

0.0074

1.000

0.0071

0.0155

1.000

0.0182

Model Visualisations

LOGISTIC

LogLoss: 0.0019 AUC: 1.000 Precision: 1.000 Recall: 1.000

Best params: {"C": 0.1}

Confusion Matrix

Feature Importance

Roc Pr Curve

Shap Summary

XGB

LogLoss: 0.0042 AUC: 1.000 Precision: 1.000 Recall: 1.000

Best params: {"learning_rate": 0.3, "max_depth": 3, "n_estimators": 400}

Confusion Matrix

Feature Importance

Roc Pr Curve

Shap Summary

SVM

LogLoss: 0.0074 AUC: 1.000 Precision: 1.000 Recall: 1.000

Best params: {"C": 100, "gamma": "scale", "kernel": "rbf"}

Confusion Matrix

Feature Importance

Roc Pr Curve

Shap Summary

RF

LogLoss: 0.0155 AUC: 1.000 Precision: 1.000 Recall: 1.000

Best params: {"max_depth": 10, "n_estimators": 400}

Confusion Matrix

Feature Importance

Roc Pr Curve

Shap Summary